Pacic

Journal of
Mathematics
TECHNIQUES FOR APPROACHING THE DUAL RAMSEY
PROPERTY IN THE PROJECTIVE HIERARCHY
Lorenz Halbeisen and Benedikt L

owe
Volume 200 No. 1 September 2001
PACIFIC JOURNAL OF MATHEMATICS
Vol. 200, No. 1, 2001
TECHNIQUES FOR APPROACHING THE DUAL RAMSEY
PROPERTY IN THE PROJECTIVE HIERARCHY
Lorenz Halbeisen and Benedikt L

owe
We dene the dualizations of objects and concepts which
are essential for investigating the Ramsey property in the rst
levels of the projective hierarchy, prove a forcing equivalence
theorem for dual Mathias forcing and dual Laver forcing, and
show that the Harrington-Kechris techniques for proving the
Ramsey property from determinacy work in the dualized case
as well.
1. Introduction.
Set theory of the reals is a subeld of Mathematical Logic mainly concerned
with the interplay between forcing and Descriptive Set Theory. One of
the motivations behind Descriptive Set Theory is the strong intuition that
simple sets of real numbers should not display irregular behaviour, or, in
other words, they should be topologically and measure theoretically nice.
In order to ll this statement with mathematical content, we should make
clear what we mean by simple and what we mean by nice. Both ques-
tions have a conventional and well known answer:
The measure of simplicity with which we categorize our sets of reals
is the projective hierarchy, in other words, the number of quantiers
necessary to dene the sets with a formula in rst order analysis (or
second order arithmetic).
A set should be considered nice or regular if it has the Baire
property in all naturally occurring topologies on the real numbers and
is a member of all conceivably natural -algebras.
Set theory teaches us that the axioms of ZFC do not entail a formal
version of these intuitions: It is consistent with ZFC that there are irregular
sets already at the rst level of the projective hierarchy.
1
Thus the focus
shifts from proving that all simple sets are nice to investigating the situations
under which our intuitions are met by the facts.
1
In Godels Constructible Universe L there is a
1
2
set which is not Lebesgue measur-
able and which does not have the Baire property. Worse still, there is an uncountable
1
1
set with no perfect subset and a
1
1
set which is not Martin measurable.
119
120 L. HALBEISEN AND B. L

OWE
A whole array of research in this direction is dealing with the second level
of the projective hierarchy. Solovay provided us with the prototype of a
characterization theorem for the second level:
Theorem 1.1. The following are equivalent:
(i) Every
1
2
set of reals has the Baire property.
(ii) For every real a []

the set of Cohen generic reals over the model

L[a] is comeager in the standard topology on the real numbers.
One could call a theorem like this a transcendence principle over the con-
structible universe. These principles connect the theory of forcing and the
topological properties of the reals. Comparable theorems have been proved
in [JuSh89] (for the
1
2
level) and in [BrLo99] (for dierent topologies and
-algebras).
A particularly interesting instance of niceness in the above sense is the
Ramsey property, a topological property which is deeply connected to
Ramsey theory and innitary combinatorics. The Ramsey property is linked
to a forcing notion called Mathias forcing, introduced by Mathias in
[Mat77], and Judah and Shelah were able to obtain the following Solovay
type characterization for it (cf. [JuSh89, Theorem 2.7 & Theorem 2.8]):
Theorem 1.2. The following are equivalent:
(i) Every
1
2
set of reals has the Ramsey property.
(ii) Every
1
2
set of reals has the Ramsey property.
(iii) For every real a []

the set r []

: r is Ramsey over L[a][T

r
]
is comeager in the Ellentuck topology.
2
One connection to Mathias forcing is given by the following result (cf.
[HalbJu96, Theorem 4.1]):
Proposition 1.3. If N is any model of ZFC, then the following are equiv-
alent:
(i) N is a model in which every
1
2
set is Ramsey, and
(ii) N is
1
3
-Mathias-absolute.
3
As the Ramsey property talks about innite subsets of the natural num-
bers, it is easily dualized by something we shall call the dual Ramsey
property, talking about innite partitions of the natural numbers.
4
This
2
A real r is Ramsey over L[a][F
r
] if and only if F
r
:= D
r
L[a][D
r
] forms an ultralter
in L[a][F
r
], where D
r
:= {r

[]

: r

}, and r is L
F
r
-generic over L[a][F
r
], where
L
F
r
is Laver forcing restricted to F
r
.
3
Similar characterizations also exist for some other properties, e.g., for Lebesgue mea-
surability and Baire property (cf. [BaJu95, Theorem 9.3.8]).
4
Innite subsets can be seen as images of injective functions and innite partitions can
be seen as preimages of surjective functions, so the move from innite subsets to innite
partitions actually is a dualization process.
APPROACHING THE DUAL RAMSEY PROPERTY 121
property has been introduced by Carlson and Simpson in [CaSi84] and
further investigated in [Halb98-2] and [Halb98-1].
One thing that is striking about the relationship between the Ramsey
property and the dual Ramsey property are the distinctive symmetries and
asymmetries. This paper can be understood as a catalogue of some of the
similarities; in fact, one could see parts of this paper as an attempt to reach
the obvious dualization of Theorem 1.2:
Conjecture 1.4. The following are equivalent:
(i) Every
1
2
set of reals has the dual Ramsey property.
(ii) Every
1
2
set of reals has the dual Ramsey property.
(iii) For every real a []

the set R : R is dual Ramsey over L[a][D

R
]
is comeager in the dual Ellentuck topology.
In order to approach this conjecture and to give an idea what dual Ram-
sey over L[a][D
R
] could mean, several of the techniques of [JuSh89] and
[Mat77] have to be adapted to the new environment:
Mathias forcing has a characteristic product form M = {()/n M
U
where U is the canonical name for the generic ultralter added by {()/n.
This ultralter is in fact a Ramsey ultralter,
5
and Judah and Shelah show
in their [JuSh89] that Mathias forcing relative to an ultralter is forcing-
equivalent to Laver forcing relative to the same ultralter, provided that the
ultralter is Ramsey (cf. [JuSh89, Theorem 1.20 (i)]):
Theorem 1.5. Let T be a Ramsey ultralter. Then the forcing notions L
F
and M
F
are equivalent.
This theorem was our motivation to search for a dual version of Laver forc-
ing and the dualization of Ramsey ultralters to work towards a dualization
of Theorem 1.2.
In our dualized situation there are many things to be done to make sense of
the dualized versions: One has to nd a dualized version of {()/n and to
prove the corresponding product form of dual Mathias forcing (already done
in [Halb98-1]), one has to nd a dualized version of Ramsey ultralters,
and one has to make explicit what Laver forcing in this context is supposed
to mean.
Section 2 of this paper denes all the dualized notions needed for the tech-
nical work on the dual Ramsey property. In Section 3, the reader will nd a
couple of facts about a dualization of Ramsey ultralters; their connection
to the game lters from [Halb98-1] is given in the appendix. Section 4
5
A set F []

is a Ramsey lter if F is a lter and for any colouring : []

n
r +1
(with n, r ) there is an x F such that [x]
n
is constant. Notice that every Ramsey
lter is an ultralter.
122 L. HALBEISEN AND B. L

OWE
moves on to discuss dual Laver forcing and proves the dualized version of
Theorem 1.5.
In Sections 5 and 6 we investigate the extent of sets with the dual Ram-
sey property in the projective hierarchy. In Section 5 we prove a couple
of consistency results for the rst three levels of the projective hierarchy.
After that, Section 6 looks at the dual Ramsey property from a completely
dierent angle: If we assume an appropriate amount of determinacy, we
know that a large collection of sets has the Ramsey property. This re-
sult is not at all immediate from the Banach-Mazur game for the topology
associated with the Ramsey property.
6
However, in that section we note
that the Harrington-Kechris technique of proving the Ramsey property from
standard determinacy (cf. [HarKe81]) alone works for the dualized case as
well.
It should be mentioned that the technicalities of the dualization process
are not always as easy as they seem in retrospect. Finding the correct and
natural dualizations for the interesting notions from the classical case is the
most challenging part in this project. After the right dualizations are at
hand, in most cases one can follow the classical proofs. So, the merits of
this paper lie mainly in the denitions that make the proofs nice and easy
and give a proper and rmly rooted understanding of the symmetries. This
is also the reason for the unproportional size of Section 2 compared to the
other sections.
2. Denitions and notations.
2.0. Set-theoretic notation. Most of our set-theoretic notation is stan-
dard and can be found in textbooks like [Je78], [Ku83] or [BaJu95]. For
the denitions and some basic facts concerning the projective hierarchy we
refer the reader to [Kan94, '12].
We shall consider the set []

as the set of real numbers. For the Turing

join of two reals x and y (i.e., coding two reals into one), we use the standard
notation x y.
2.1. Partitions. A set P {(S) is a partition of the set S if / A,

P = S and for all distinct p

1
, p
2
P we have p
1
p
2
= . An element of
a partition P is also called a block of P and dom(P) :=

P is called the
domain of P. A partition P is called innite, if [P[ is innite, where [P[
denotes the cardinality of the set P. The equivalence relation on S uniquely
determined by a partition P is denoted by
P
.
6
The obstacle is that playing basic open sets in this topology cannot be coded by
natural numbers. So the Banach-Mazur games essentially needs determinacy for games
with real moves, e.g., PD
R
. This is connected to the famous open question whether AD
implies that every set has the Ramsey property (cf. [Kan94, Question 27.18]).
APPROACHING THE DUAL RAMSEY PROPERTY 123
Let P and Q be two arbitrary partitions. We say that P is coarser than
Q (or that Q is ner than P) and write P _ Q, if for all blocks p P, the
set p dom(Q) is the union of some sets q
i
dom(P), where each q
i
is a
block of Q. Let P Q be the nest partition which is coarser than P and Q
with dom(P Q) = dom(P) dom(Q). We say that P is almost coarser
than Q and write P _

Q if there is a partition R such that dom(R) is nite

and R P _ Q. If P _

Q and Q _

P, then we write P

= Q.
7
Let P and Q be two partitions. If for each p P there is a q Q
such that p = q dom(P), we write P Q. Note that P Q implies
dom(P) dom(Q).
For x let min(x) :=

x. If P is a partition with dom(P) , then
Min(P) := min(p) : p P; and for n , P(n) denotes the unique block
p P such that [ min(p) Min(P)[ = n + 1.
The set of all innite partitions of is denoted by ()

; and the set of

all partitions s with dom(s) is denoted by (N).
For s (N), let s

denote the partition s

dom(s)

. Notice that
[s

[ = [s[ + 1.
For a natural number n, let ()
n
denote the set of all u (N) such that
[u[ = n. Further, for n and X ()

let
(X)
n
:= u (N) : [u[ = n u

_ X ,
and for s (N) such that [s[ n and s _ X, let
(s, X)
n
:= u (N) : [u[ = n s u u

_ X .
It will be convenient to consider as the partition which contains only
singletons, and therefore, for s (N), (s, )
n
:= u (N) : [u[ = ns u.
A family F ()

is called a lter if
() / F;
() If X F and X _ Y , then Y F;
() If X and Y belong to F, then X Y F.
Further, we call F ()

an ultralter if F is a lter which is not properly

contained in any lter. Notice that if X F and F ()

is an ultralter,
then each Y ()

with Y

= X belongs to F, too.
2.2. The dual Ellentuck topology and the dual Ramsey property.
Let X ()

and s (N) be such that s _ X. Then

(s, X)

:= Y ()

: s Y _ X
and
(X)

:= (, X)

= Y ()

: Y _ X.
7
We choose this notation because the properties of and are similar to those of
and .
124 L. HALBEISEN AND B. L

OWE
Obviously, this denition depends on the model we are working in, so, if
this should become important, we denote by (s, X)

N
the corresponding set
interpreted in the model N.
Let the basic open sets on ()

be and the sets (s, X)

, where s and X
are as above. These sets are called the dual Ellentuck neighbourhoods.
The topology induced by the dual Ellentuck neighbourhoods is called the
dual Ellentuck topology (cf. [CaSi84]).
A family A ()

has the dual Ramsey property (or just is dual

Ramsey) if and only if there is a partition X ()

such that either

(X)

A or (X)

A = .
Closely related (but stronger) is the notion of a completely dual Ramsey
set: A set A ()

is said to be completely dual Ramsey if and only

if for each dual Ellentuck neighbourhood (s, X)

there is a Y (s, X)

such that (s, Y )

A or (s, Y )

A = . If we are always in the latter

case, then A is called completely dual Ramsey-null. It is not clear if
every projective set is completely dual Ramsey is really stronger than just
every projective set is dual Ramsey, because we cannot simply translate
Lemma 2.1 of [BrLo99], where it is shown among other things that every
projective set is Ramsey and every projective set is completely Ramsey
are equivalent.
Carlson and Simpson proved in [CaSi84] that a set A is completely dual
Ramsey if and only if A has the Baire property with respect to the dual
Ellentuck topology and A is completely dual Ramsey-null if and only if A is
meager with respect to the dual Ellentuck topology.
8
As a matter of fact
we like to mention that in the dual Ellentuck topology every meager set is
nowhere dense and hence, the dual Ellentuck topology is a Baire topology
(i.e., no open set is meager). This corresponds to the similar facts about
being completely Ramsey and the Ellentuck topology.
9
2.3. Dual Mathias forcing. The conditions of the dual Mathias forcing
M

= 'M

, ` are the pairs 's, X` such that (s, X)

is a non-empty dual
Ellentuck neighbourhood, and the partial order is dened by
's, X` 't, Y ` (s, X)

(t, Y )

.
If 's, X` is an M

-condition, then we call s the stem of the condition.

If G is M

-generic over N, then G induces in a canonical way an innite

partition X
G
()

such that N[G] = N[X

G
], and therefore we consider
8
A set S has the Baire property if there is an Borel set B such that the symmetric
dierence SB is meager, where a meager set is the union of countably many nowhere
dense sets.
9
Cf. [El74] & [Ke95, 19.D].
APPROACHING THE DUAL RAMSEY PROPERTY 125
the partition X
G
as the generic object. We can reconstruct the original G
from X
G
by observing that
's, X` G X
G
(s, X)

N[G]
.
Since the dual Ellentuck topology is innately connected with dual Mathias
forcing, we choose the following notation for meager and comeager sets in
the dual Ellentuck topology:
A (m

0
) A is dual Ellentuck meager, and
A (m

1
) A is dual Ellentuck comeager,
i.e., ()

` A is dual Ellentuck meager.

Since A is dual Ellentuck meager if and only if A is completely dual
Ramsey-null, (m

0
) {(()

) is also the ideal of completely dual Ramsey-

null sets.
The following fact gives two properties of dual Mathias forcing which also
hold for Mathias forcing.
Fact 2.1. If X
G
is M

-generic and Y (X
G
)

, then Y is M

-generic as
well (we will call this property the homogeneity property); and therefore,
dual Mathias forcing is proper. Moreover, for any sentence of the forcing
language M

and for any M

-condition 's, X`, there is an M

-condition
's, Y ` 's, X` such that 's, Y `
M
or 's, Y `
M
(this property is
called pure decision).
Proof. For a proof, cf. [CaSi84, Theorem 5.5 & Theorem 5.2].
As an immediate consequence we get that the set of dual Mathias generic
partitions over every model N is either empty or a non-meager set which is
completely dual Ramsey.
Like Mathias forcing, dual Mathias forcing has also a characteristic prod-
uct form.
Let U

= '()

, ` be the partial order dened as follows:

X Y X _

Y.
U

is the natural dualization of {()/n.

For a family E ()

we dene the restricted dual Mathias forcing

M

E
as follows. The conditions of M

E
= 'M

E
, ` are the M

-conditions
's, X` such that X E.
Now we get
Fact 2.2. M

= U

G
, where Gis the canonical name for the U

-generic
object.
Proof. For a proof, cf. [Halb98-1, Fact 2.5].
126 L. HALBEISEN AND B. L

OWE
2.4. Restricted dual Laver forcing. In order to dene the forcing notion
which will be investigated later on, we rst have to give some notations.
For T (N) and t T we dene the successor set of t in T as follows:
succ
T
(t) := u T : t u [u[ = [t[ + 1 .
Let E ()

be any non-empty family (later on we investigate only the

case when E is an ultralter).
With respect to E, we dene the dual Laver forcing restricted to E,
denoted by L

E
= 'L

E
, `, as follows:
() p L

E
if and only if p (N) with the property that there is an s p
(denoted stem(p)) such that for all t p we have s t.
() There exists a set X
p
t
: t p E such that for t p we have t

_ X
p
t
and
succ
p
(t) = u : u (t

, X
p
t
)
(|t|+1)
.
Further, for t, u p with t u we have
(u, X
p
u
)

(t, X
p
t
)

,
and if dom(t) = dom(u) and t _ u, then
X
p
t
= X
p
u
.
() For two L

E
-conditions p and q we stipulate
p q p q.
Notice that p q implies stem(q) stem(p) and hence, if G L

E
is L

E
-
generic over some N, then the set s : s = stem(p) for some p G forms
in a canonical way a partition X
G
()

. Moreover, N[G] = N[X

G
] and
therefore we may consider also the partition X
G
as the L

E
-generic object.
For an L

E
-condition p we call a partition X ()

a branch of p if each
t (N) with t

_ X belongs to p.
Fact 2.3. If X is a branch of the L

E
-condition p where stem(p) = s and
Y (s, X)

, then Y is a branch of p, too.

Proof. This follows immediately from ().
2.5. Special ultralters on ()

. A family F has the segment colour-

ing property (or just scp) if for any s _ X F with [s[ = n and for
any colouring : (s, X)
(n+k)
r, where r and n + k are positive natural
numbers, there is a Y (s, X)

F such that (s, Y )

(n+k)
is monochromatic.
A family F ()

is an scp-lter if F is a lter which has the segment

colouring property.
In Section 7 we shall introduce the notion of game lters (from
[Halb98-1]) and show that game lters are scp-lters.
Fact 2.4. If F ()

is an scp-lter, then F is an ultralter.

APPROACHING THE DUAL RAMSEY PROPERTY 127
Proof. Let F ()

be an scp-lter and assume that there exists an X

()

such that for every Y F, X Y ()

. Let : ()
n
2 be such
that (s) = 0 if s (X)
n
, otherwise (s) = 1. Because F has the segment
colouring property, we nd a Y F such that (Y )
n
is constant. If
(Y )
n
= 1, then XY / ()

which contradicts the assumption. Thus,

(Y )
n
= 0, which implies X F and hence, the lter F is maximal.
A family F is diagonalizable if for any L

F
-condition p, there is a partition
X F such that X is a branch of p. Notice that a diagonalizable family can
also be characterized by a two player game, where the L

F
-condition p can
be considered as a strategy for player I.
A family F is a Ramsey

lter if F is a diagonalizable scp-lter.

In Footnote 5 we have dened Ramsey ultralters over in terms of
colourings. This denition corresponds to the denition of scp-lters. On
the other hand, Galvin and Shelah proved that Ramsey ultralters can be
characterized as well by a two player game without a winning strategy for
player I, where a winning strategy for player I is in fact a restricted Laver-
condition (cf. [BaJu95, Theorem 4.5.3]). This denition of Ramsey ultra-
lters corresponds to diagonalizable lters. It is possible that the notions
of scp-lters and diagonalizable lters are equivalent, but this is still
open.
Beyond the dualization of the notion of a Ramsey ultralter, the dual-
ization process leading from []

to ()

has interesting consequences for

the spaces of ultralters on these spaces. These consequences belong to the
asymmetrical aspects of the relationship between []

and ()

and are the

point of focus in [HalbLo].
2.6. Switching between reals and partitions. We x : []
2
to
be any arithmetic bijection between the set of pairs of natural numbers and
.
Let x []

; then the set trans(x) is dened by

n trans(x) : s
<

n = (s(0), s([s[ 1)) and

k [s[ 1((s(k), s(k + 1)) x)

.
As the name suggests, trans(x) is the set of codes of pairs in the transitive
closure of the relation (k, ) x. A real x is called transitive if trans(x) =
x.
Note that in general trans(x) x and that the relation
R
x
(k, ) : (k, ) trans(x)
128 L. HALBEISEN AND B. L

OWE
is symmetric (by choice of the domain of ) and transitive. Thus, if x []

,
we can consider x as a partition (by reexivization of R
x
) via
n
x
m : n = m or (n, m) trans(x).
We call this partition the corresponding partition of x []

, and
denote it by cp(x). Note that cp(x) ()

if
kn > km < n((n
x
m))
and further if y x, then cp(y) _ cp(x).
We encode a partition X of by a real pc(X) (the partition code of
X) as follows.
pc(X) := k : nm(k = (n, m) (n
X
m)).
Note that if X _ Y then pc(X) pc(Y ).
Notice that both the function pc and the function cp are arithmetic, and
that they are in a sense inverse to each other:
Observation 2.5. For every X ()

and every x []

the following
hold:
(i) cp(pc(X)) = X, and
(ii) if x is transitive, then pc(cp(x)) = x.
Now, a set A []

has the dual Ramsey property (or just is dual

Ramsey) if and only if the set X ()

: x A(X = cp(x)) has the

dual Ramsey property. By Observation 2.5, this is equivalent to saying that
the set X ()

: pc(X) A has the dual Ramsey property.

By the denition of the dual Ramsey property we have that every
1
n
set
is dual Ramsey if and only if every
1
n
set is dual Ramsey. Furthermore, we
have by [Halb98-1, Lemma 7.2] that if every
1
n
set is dual Ramsey then
every
1
n
set has the classical Ramsey property.
As a matter of fact we like to mention the following
Proposition 2.6. If every
1
n
set has the dual Ramsey property, then every

1
n
set has the Ramsey property.
Proof. Suppose A is a
1
n
set of reals. Let be a
1
n
formula and be a

1
n
formula witnessing this, i.e.,
x A (x) (x).
To show that A is Ramsey we dene a dierent
1
n
set by formulae

and

as follows:

(v) : w(w = Min(cp(v)) (w)),

(v) : w(w = Min(cp(v)) (w)).

APPROACHING THE DUAL RAMSEY PROPERTY 129
Obviously,

is
1
n
and

is
1
n
, and since Min(cp(v)) is uniquely
determined for each v, these two formulae are equivalent and hence dene
a
1
n
set A

of reals. The rest of the proof is exactly as in [Halb98-1,

Lemma 7.2].
And as a corollary we get
Corollary 2.7. If every
1
2
set is dual Ramsey, then every
1
2
set is Ram-
sey.
Proof. This follows immediately from Proposition 2.6 by Theorem 1.2.
2.7. Descriptive Set Theory of the Cabal. For our results in Section 6
we shall need some basic notions of the Descriptive Set Theory of the Cabal
Seminar. Everything we lay out here can be found in [Mo80], our account
is just for the convenience of the more combinatorially oriented reader who
might be unfamiliar with the language of the Cabal.
We shall presuppose basic knowledge with the standard notation for de-
terminacy and the elementary results of the theory of perfect information
games as outlined in [Kan94, '27].
Let X be a set of reals and Ord. Any surjective function : X
is called a norm on X. The ordinal is called the length of . A family
:= '
n
: n ` of norms on X is called a scale on X if for every sequence
'x
i
: i ` X and every n the following holds: If '
n
(x
i
) : i ` is
eventually constant, say, equal to
n
, then x := lim
i
x
i
X and
n
(x)

n
.
10
Let be any pointclass, any norm on X, and any scale on X. We
shall call a norm if there are two relations R and R

in such that:
y X x

(x X (x) (y)) R(x, y) R

(x, y)

.
We call a scale a scale if all norms
n
occurring in are norms,
uniformly in n.
11
We shall say that a set X admits a norm (a scale)
if there is a norm (a scale) on X that is a norm (a scale).
The fundamental theorems connecting determinacy, norms and scales are
the Periodicity Theorems of [AdMo68], [Mar68] and [Mo71]. In the
following we shall need the rst two Periodicity Theorems in special cases:
First Periodicity Theorem 2.8. Suppose that Det(
1
2n
) holds and x
[]

is a real. Then every

1
2n+1
(x) set admits a
1
2n+1
(x) norm.
Second Periodicity Theorem 2.9. Suppose that Det(
1
2n
) holds and x
[]

is a real. Then every

1
2n+1
(x) set admits a
1
2n+1
(x) scale.
10
For the basic theory of scales, cf. [KeMo78].
11
A more precise denition can be found in [Mo80, p. 228].
130 L. HALBEISEN AND B. L

OWE
For proofs of these theorems (in a much more general formulation), we
refer the reader to [Mo80, 6B.1 & 6C.3].
We dene (for every n ) the projective ordinals by

1
n
:= sup|| : is a
1
n
prewellordering on []

,
and note that for every
1
2n+1
complete set the length of every
1
2n+1
norm
on it is exactly
1
2n+1
([Mo80, 4C.14]).
Let P
x
2n+1
be a
1
2n+1
(x) complete set of reals. Assuming Det(
1
2n
) we
get a
1
2n+1
(x) scale
x
=
x
m
: m for P
x
2n+1
by Theorem 2.9.
For any real y P
x
2n+1
, we denote by
x
(y) the sequence of ordinals
determined by the scale, i.e.,
x
(y) = '
x
m
(y) : m `.
Now let
T
x
2n+1
:= 'ym,
x
(y)m` : y P
x
2n+1
, m
be the tree associated to
x
. By the remark about the lengths of norms,
it is a tree on
1
2n+1
. If x is any recursive real, we write T
2n+1
instead
of T
x
2n+1
.
The model L[T
2n+1
] can be seen as an analogue of the constructible uni-
verse L in the odd projective levels: The (Shoeneld)
1
1
scale for a
1
1
complete set is in L, hence L[T
1
] = L.
12
Indeed, the reals of L[T
2n+1
] are
exactly the reals of M
2n
, the canonical iterable inner model with 2n Woodin
cardinals.
13
Moreover, not just the reals of the models, but the models L[T
2n+1
] them-
selves are independent of the choices of the particular
1
2n+1
complete set
and the scale on it, as has been shown by Becker and Kechris in [BeKe84,
Theorem 1 & 2]:
Theorem 2.10. Assume PD and let x []

be a real. If P and Q are

1
2n+1
(x) complete sets, and are scales on P and Q, respectively, and
T and S are the trees associated to and , respectively. Then L[T] = L[S].
Another consequence of determinacy which will be mentioned only briey
to simplify notation is the existence of largest countable sets of certain (light-
face) complexity classes:
Theorem 2.11. Let x []

be a real. Suppose that Det(

1
2n
(x)) holds.
Then there is a largest countable
1
2n+2
(x) set which will be denoted by
C
2n+2
(x).
14
Proof. Cf. [KeMo72, Theorem 2].
12
For a proof, cf. [KeMo78, 9C].
13
Combine [St95, Corollary 4.9] with [HarKe81, Theorem 7.2.1].
14
As for the trees T
2n+1
, we shall omit the parameter x and write C
2n+1
if the real x
is recursive.
APPROACHING THE DUAL RAMSEY PROPERTY 131
3. On Ramsey

ultralters.
In this section we show that Ramsey

ultralters exist if we assume CH and

that in general both existence and non-existence of Ramsey

ultralters are
consistent with ZFC.
First we show that an scp-ultralter induces in a canonical way a Ramsey
lter on .
Fact 3.1. If F ()

is an scp-ultralter, then Min(X) : X F ` 0 is

a Ramsey lter on .
Proof. For positive natural numbers n and r let : []
n
r be any colour-
ing. We dene : ()
n
r by stipulating (s) :=

Min(s

) ` 0

. It is
easy to see that if (X)
n
is constant for an X F, then [Min(X) `0]
n
is constant, too.
Proposition 3.2. It is consistent with ZFC that there are no scp-ultralters.
Proof. Kunen proved (cf. [Je78, Theorem 91]) that it is consistent with
ZFC that there are no Ramsey lters on . Therefore, by Fact 3.1, in a
model of ZFC in which there are no Ramsey lters, there are also no scp-
ultralters.
Let U

= '()

, ` be the partial order dened as in Subsection 2.3. It

is easy to see that the forcing notion U

is -closed (this is part of Fact 2.3

of [Halb98-1]).
Lemma 3.3. If G is U

-generic over V, then G is an scp-ultralter in

V[G].
Proof. Let s (N) and k with [s[ = n and n + k > 0. Further, let
: ()
(n+k)
r be any colouring and for s _ X ()

let
H
(s,X)
:= Y (s, X)

: (s, X)(s, Y )
(n+k)
is constant .
By the main result of [Halb] and its proof, for every dual Ellentuck neigh-
bourhood (s, X)

and for any colouring : (s, X)

(n+k)
r, there is a
Y (s, X)

such that (s, Y )

(n+k)
is constant. Hence, for any dual El-
lentuck neighbourhood (s, X)

and for any colouring : ()

(n+k)
r,
the set H
(s,X)
is dense below X. Because every such colouring can be
encoded by a real and U

is -closed, the forcing notion U

does not add

any colouring , which implies, because G meets each dense set, that G is
an scp-ultralter in V[G].
We can prove with similar arguments:
Lemma 3.4. If G is U

-generic over V, then G is a diagonalizable ultra-

lter in V[G].
132 L. HALBEISEN AND B. L

OWE
Proof. Let p be a U

-name such that

U
p is an L

G
-condition,
where G is the canonical name for the U

-generic object, and let X be

any U

-condition. Because p can be encoded by a real number and U

is -closed, there is a U

-condition Y X and a real p

V such that
Y
U
p

= p, which implies Y _

succ
p
(t) for every t p

. By induction
one can construct a Z _

Y such that Z is a branch of p

and therefore,
Z
U
there is a branch of p which belongs to G.
Since Z X, this completes the proof.
Proposition 3.5. Assume CH, then there is a Ramsey

ultralter.
Proof. Assume V [= CH. Let be large enough such that {(()

) H(),
i.e., the power set of ()

(in V) is hereditarily of size < . Let N be an

elementary submodel of 'H(), ` containing all reals of V with [N[ = 2

0
.
We consider the forcing notion U

in the model N. Because [N[ = 2

0
, in
V there is an enumeration D

()

: < 2

0
of all dense sets of U

which lie in N. Since U

is -closed and because V [= CH, U

is 2

0
-closed
in V and therefore we can construct a descending sequence p

: < 2

in V such that p

for each < 2

0
. Let G := p ()

: p

_
p for some p

, then G is U

-generic over N. By Lemma 3.3 and Lemma 3.4

we have N[G] [= there is a Ramsey

ultralter, and because N contains

all reals of V and every function f : ()
n
r (where n, r ) and every
L

G
-condition p can be encoded by a real number, the Ramsey

ultralter
in N[G] is also a Ramsey

ultralter in V, which completes the proof.

4. On L

F
and M

F
for Ramsey

lters F.
In this section, F ()

denotes always a Ramsey

ultralter.
We shall show that the forcing notions L

F
and M

F
are equivalent and
that both forcing notions have pure decision and the homogeneity property
(this means that coarsenings of generic objects remain generic, see Fact 2.1).
We show rst that M

F
has pure decision and the homogeneity property. To
show this we will follow [Halb98-1, Section 4].
If s (N) and s _ X F, then we call the dual Ellentuck neighbourhood
(s, X)

an F-dual Ellentuck neighbourhood and write (s, X)

F
to em-
phasize that X F. A set O ()

is called F-open if O can be written

as the union of some F-dual Ellentuck neighbourhoods.
For s (N) remember that s

= s

dom(s)

.
Let O ()

be an F-open set. Call (s, X)

F
good (with respect to O),
if for some Y (s, X)

F
F, (s, Y )

F
O; otherwise call it bad. Note that if
(s, X)

F
is bad and Y (s, X)

F
F, then (s, Y )

F
is bad, too. We call (s, X)

F
APPROACHING THE DUAL RAMSEY PROPERTY 133
ugly if (t

, X)

F
is bad for all s t

_ X with [t[ = [s[. Note that if (s, X)

F
is ugly, then (s, X)

F
is bad.
Lemma 4.1. Let F ()

be a Ramsey

ultralter and O ()

an F-
open set. If (s, X)

F
is bad (with respect to O), then there is a Z (s, X)

F
such that (s, Z)

F
is ugly.
Proof. We begin by constructing an L

F
-condition p. Let s
0
be such that
s s

0
_ X and [s[ = [s
0
[, and put stem(p) := s
0
. If there is an Y
(s

0
, X)

F
F such that (s

0
, Y )

F
O, then X
s
0
:= Y , otherwise, X
s
0
:= X.
Let s

n+1
(s
n
X
s
n
) be such that [s
n+1
[ = [s
n
[ + 1 = [s[ + n + 1 and let
t
i
: i h be an enumeration of all t such that s
0
t _ s
n+1
, [t[ = [s[ and
dom(t) = dom(s
n+1
). Further let Y
1
:= X
s
n
. Now choose for each i h a
partition Y
i
F such that Y
i
_ Y
i1
, s

n+1
Y
i
and ((t
i
)

, Y
i
)

F
is bad or
((t
i
n
)

, Y
i
)

F
O and nally, let X
s
n+1
:= Y
h
.
Put p := t (N) : s
0
t

_ s

n
for some n . Since F is diagonaliz-
able, there is a partition Y F which is a branch of p. We may assume that
s
0
Y . Dene S
Y
:= t : s t

_ Y [t[ = [s[; then, by the construction

of p, for all t S
Y
we have either (t

, Y )

F
is bad or (t

, Y )

F
O. Now let
B
0
:= t S
Y
: (t, Y )

F
is bad and B
1
:= t S
Y
: (t

, Y )

F
O = S
Y
`B
0
.
Because F is an scp-lter, there is a partition Z (s, Y )

F
F such that
S
Z
B
0
or S
Z
B
1
. If we are in the latter case, we have (s, Z)

F
O,
which is a contradiction to our assumption that (s, X)

F
is bad. So, we must
have S
Z
B
0
, which implies that (s, Z)

F
is ugly and completes the proof
of the Lemma.
Lemma 4.2. If F is a Ramsey

ultralter and O ()

is an F-open
set, then for every F-dual Ellentuck neighbourhood (s, X)

F
, there is a Y
(s, X)

F
F such that (s, Y )

F
O or (s, Y )

F
O F = .
Proof. If (s, X)

F
is good, then we are done. Otherwise, we can construct
an L

F
-condition p in a similar way as in Lemma 4.1, such that for any
branch Y of p which belongs to F we have the following: For each t with
s t

_ Y , the set (t

, Y )

F
is bad. We claim that (s, Y )

F
OF = . Take
any Z (s, Y )

F
O F. Because O is F-open we nd a t Z such that
(t

, Z)

F
O. Because s t

_ Y we have by construction that (t

, Y )

F
is
bad. Hence, there is no Z (t

, Y )

F
such that (t

, Z)

F
O. This completes
the proof.
Now we can show that M

F
has pure decision and the homogeneity prop-
erty.
Theorem 4.3. Let F be a Ramsey

ultralter and let be a sentence of

the forcing language. For any M

F
-condition 's, X` there is a M

F
-condition
's, Y ` 's, X` such that 's, Y `
M

F
or 's, Y `
M

F
.
134 L. HALBEISEN AND B. L

OWE
Proof. The proof is same as the proof of [Halb98-1, Theorem 4.3], using
Lemma 4.2.
The next theorem shows in fact that if F is a Ramsey

ultralter, then
M

F
is proper.
Theorem 4.4. Let F ()

be a Ramsey

ultralter, then M

F
has the
homogeneity property.
Proof. The proof is same as the proof of [Halb98-1, Theorem 4.4], using
Lemma 4.2.
In order to show that M

F
and L

F
are equivalent if F is a Ramsey

ultra-
lter, we dene rst some special L

F
-conditions.
An L

F
-condition p is called uniform if there is a partition X F such
that (t, X)

= (t, X
p
t
)

for every t p; this partition is denoted by u(p).

These conditions roughly correspond to the simple conditions of [JuSh89,
Denition 1.10].
Lemma 4.5. If F is a Ramsey

ultralter, then the set of all uniform L

F
-
conditions is dense and open in L

F
.
Proof. Let p L

F
with s = stem(p), then, since F is diagonalizable, there
is an X F which is a branch of p. Let q be the uniform condition with
u(q) = X and stem(q) = s. Note that X is a branch of q. By Fact 2.3, each
Y (s, X)

is also a branch of p, which implies that q p.

Theorem 4.6. If F is a Ramsey

ultralter, then M

F
and L

F
are forcing
equivalent.
Proof. Let I := p L

F
: p is uniform and dene
j : I M

F
p 'stem(p), u(p)` ,
then it is easily checked that j is a dense embedding and because (by
Lemma 4.5) I is dense open in L

F
, this completes the proof.
This is the promised dualization of Theorem 1.5 and possibly one step
towards a proof of Conjecture 1.4.
5. The dual Ramsey property for simple pointclasses.
In the following we will show that it is consistent with ZFC that the sets in
the rst levels of the projective hierarchy are dual Ramsey. We begin with
the analytic sets:
Because M

has pure decision and the homogeneity property, one can

show the pretty straightforward
APPROACHING THE DUAL RAMSEY PROPERTY 135
Fact 5.1. Analytic sets are completely dual Ramsey.
Proof. Let A be an arbitrary
1
1
(a) set with parameter a []

and let
(s, Y )

be any dual Ellentuck neighbourhood and 's, Y ` the corresponding

M

-condition. Take a countable model N of a suciently large fragment

of ZFC which contains Y and a. Let X
G
be the canonical name for the
M

-generic object. Because M

has pure decision we nd an M

-condition
's, Z` 's, Y ` which decides X
G


A. Since N is countable, there is an
X (s, Z)

which is M

-generic over N and because every X

(s, X)

is
also M

-generic we have
N[X] [= (s, X)

A or (s, X)

A = .
Because A and (s, Y )

were arbitrary and

1
1
sets are absolute between V
and N, we are done.
Note that Fact 5.1 is veried without any reference to forcing by looking
at the unfolded version of the BanachMazur game for the dual Ellentuck
topology.
15
Remember that according to Proposition 2.6 if every
1
2
set is dual Ram-
sey then every
1
2
set has the classical Ramsey property. Because it is not
provable in ZFC that every
1
2
set is Ramsey, it is also not provable in ZFC
that every
1
2
set is dual Ramsey (e.g., L [= There is a
1
2
set which is
not dual Ramsey). On the other hand we have
Fact 5.2. The following theories are equiconsistent:
(a) ZFC.
(b) ZFC + CH + every
1
2
set is dual Ramsey.
(c) ZFC + 2

0
=
2
+ every
1
2
set is dual Ramsey.
Proof. We get both (b) and (c) by an iteration (of length
1
and
2
respec-
tively) of dual Mathias forcing with countable support, starting from L (cf.
also [Halb98-1, Theorems 6.2 & 6.3]).
Remember that (m

0
) = A ()

: A is completely dual Ramsey-null

and let
add(m

0
) := min

[E[ : E (m

0
)

E / (m

0
)

and
cov(m

0
) := min

[E[ : E (m

0
)

E = ()

.
In [Halb98-2] it is shown that add(m

0
) = cov(m

0
) = H, where H is the
dual shattering cardinal. If (m
0
) denotes the ideal of classical completely
Ramsey-null sets, then we get the analogous result, namely add(m
0
) =
cov(m
0
) = h, where h is the shattering cardinal (cf. [Pl86]). Because every
15
Cf. [Ke95, Theorem (21.8)].
136 L. HALBEISEN AND B. L

OWE

1
2
set is the union of
1
Borel sets (cf. [Je78, Theorem 95]), it is easy to see
that H >
1
implies that every
1
2
set is even completely dual Ramsey (and
the analogous result holds for the classical Ramsey property with respect
to h). Now, an
2
-iteration with countable support of dual Mathias forcing
starting from L yields a model in which H =
2
(cf. [Halb98-2]). Thus,
this provides another proof that Every
1
2
set is dual Ramsey is consistent
with ZFC. In Section 6 we shall provide a third proof as a byproduct of the
analysis of scales under PD.
Concerning Martins AxiomMA, it is well-known that MA implies h = 2

0
.
Hence, by the facts mentioned above, MA + CH implies that all
1
2
sets
have the classical Ramsey property.
A similar argument for the dualized case does not work: Brendle has
shown in [Br00-2], that MA + 2

0
> H =
1
is consistent with ZFC. How-
ever, this result does not preclude that MA + CH might imply that every

1
2
set is dual Ramsey.
At the next level we like to mention the following
Fact 5.3. Every
1
3
set is dual Ramsey is consistent with ZFC.
Proof. An
1
-iteration (or also an
2
-iteration) with countable support of
dual Mathias forcing starting from L yields a model in which every
1
3
set is
dual Ramsey. The proof is exactly the same as the proof of the corresponding
result for the classical Ramsey property given in [JuSh93]; the reason for
this is that they needed only that Mathias forcing is proper and has the
homogeneity property, but these two properties hold also for dual Mathias
forcing.
The
1
3
level is probably as far as we can get in ZFC without further as-
sumptions. It is a famous open question whether the consistency of Every

1
3
set is Ramsey implies the existence of an inner model with an inacces-
sible cardinal (cf. [Kan94, Question 11.16] & [Rai84, p. 49]). Most likely,
the dualized question is equally hard to conquer.
6. Determinacy and the dual Ramsey property.
We shall move on to arbitrary projective sets in this section. As we men-
tioned earlier, this means that we probably have to go beyond ZFC.
In [CaSi84, Section 5], the authors prove in fact that in the Solovay model
constructed by collapsing an inaccessible cardinal to
1
every projective set
is dual Ramsey. As we remarked, it is unknown whether the inaccessible
cardinal is necessary for that.
But there is another question connected to the dual Ramsey property of
projective sets: As with the standard Ramsey property we can ask whether
APPROACHING THE DUAL RAMSEY PROPERTY 137
an appropriate amount of determinacy implies the dual Ramsey property.
As usually with regularity properties of sets of reals we would expect that
Det(
1
n
) implies the dual Ramsey property for all
1
n+1
sets. But a direct
implication using determinacy is not as easy as with the more prominent
regularity properties (as Lebesgue measurability and the Baire property)
since the games connected to the dual Ramsey property (the BanachMazur
games in the dual Ellentuck topology) cannot be played using natural num-
bers.
The same problem had been encountered with the standard Ramsey prop-
erty and had been solved in [HarKe81] by making use of the scale property
and the Periodicity Theorems 2.8 and 2.9:
Theorem 6.1. If Det(
1
2n+2
), then every
1
2n+2
set is Ramsey.
The main ingredient of this proof was an analysis of the models L[T
2n+1
]
under Determinacy assumptions (Lemma 6.3). In the following we shall give
a brief review of the result with sketches of an adaptation to our context.
Lemma 6.2. Let be any -parametrized pointclass. If U is universal
for , A , N any model, and T N a tree such that p[T] = U. Then
there is a tree S N such that p[S] = A.
Proof. This is basically [Kan94, Proposition 13.13 (g)], apart from the as-
sertion that S N. But this is clear since the reduction function reducing
A to U is just the trivial function x 'n
0
, x` (where n
0
is the index of A
in U) and hence in N.
Lemma 6.3. Assume Det(
1
2n+2
). Then []

L[T
2n+1
] = C
2n+2
. In
particular, this is a countable set.
Proof Sketch. We shall very roughly sketch the argument of [HarKe81,
Theorem 7.2.1]:
First of all []

L[T
2n+1
] is easily seen to be
1
2n+2
. That every countable

1
2n+2
set of reals is a member of L[T
2n+1
] follows directly from Manselds
Theorem (cf. [Kan94, Theorem 14.7]) and Lemma 6.2.
16
So, what is left to show is that []

L[T
2n+1
] actually is countable. The
proof uses the following steps:
(i) Fix a
1
2n+1
norm

: P
2n+1

1
2n+1
which exists according to
Theorem 2.8.
(ii) Using

, code the tree T

2n+1
by some A
1
2n+1
and show that
L[T
2n+1
] = L[A].
(iii) For arbitrary subsets X
1
2n+1
, dene X

:= z []

(z)
X.
16
Note that by Theorem 2.10 the choice of the complete set for the denition of the
L[T
2n+1
] doesnt matter.
138 L. HALBEISEN AND B. L

OWE
(iv) Show: If X

1
2n+2
, then the set []

L[X] is contained in a
countable
1
2n+2
set.
(v) Compute: A

1
2n+2
.

Harrington and Kechris used this result to receive results about projec-
tive sets from PD alone that formerly could only be derived from stronger
hypotheses. The results for the classical Ramsey property follows Solovays
argument for the
1
2
case. We shall outline this argument in full generality
and then apply it to the dual Ramsey property.
At rst we need to relativize Lemma 6.3 in two dierent parameters:
Lemma 6.4. Assume Det(
1
2n+2
). Let x, y []

be any real numbers.

Then []

L[T
y
2n+1
, x] = C
2n+2
(x y).
Proof Sketch. As an immediate relativization of of Lemma 6.3 (for the
pointclass
1
2n+1
(y) instead of
1
2n+1
), we get:
[]

L[T
y
2n+1
] = C
2n+2
(y).
To show that []

L[T
y
2n+1
, x] is a countable set, we have to relativize
(iv) and (v) again. The obvious relativization of (iv) is
(iv*) If X

1
2n+2
(x y), then the set []

L[X] is contained in a
countable
1
2n+2
(x y) set.
Since L[T
y
2n+1
] = L[A] for some A
1
2n+1
such that A

is
1
2n+2
(y)
according to (ii) and (v), we know that L[T
y
2n+1
, x] = L[A, x]. Thus we have
to nd a set B
1
2n+1
such that L[B] = L[A, x] and B

1
2n+1
(x y).
This would prove the theorem.
The natural choice for B is:
B :=


1
2n+1
:

A

or

< n( = 2n n A)

or

< n( = 2n + 1 n x)

.
Obviously, B L[A, x] and A and x L[B], so L[A, x] = L[B]. Thus,
what is left is to show that B

is
1
2n+2
(xy). But this is easy to see using
the denition of B, and the facts that

was a
1
2n+1
norm and that A

was
1
2n+2
(y).
The following lemma is an obvious generalization of Shoenelds Abso-
luteness Lemma (cf. also [Mo80, Theorem 8G.10]):
Lemma 6.5.
1
2n+2
(x) formulae are absolute for models containing T
x
2n+1
,
i.e., if N is a model with T
x
2n+1
N and is a
1
2n+2
formula, then
X N (N [= [X, x] V [= [X, x]).
APPROACHING THE DUAL RAMSEY PROPERTY 139
Proof. By Theorem 2.10, we can assume that T
x
2n+1
was constructed using
an -universal set for
1
2n+1
(x), enabling us to use Lemma 6.2.
Thus every
1
2n+1
(x) set is represented by a tree S N. We easily get a
tree S

for each
1
2n+2
(x) set (cf. [Kan94, Proposition 13.13 (d)]).
But now the theorem follows from standard absoluteness of illfoundedness
as in Shoenelds proof (cf. [Kan94, Exercise 12.9 (a)]).
Theorem 6.6. Let x []

be a real. Suppose that there is a dual Mathias

generic partition over L[T
x
2n+1
]. Then every
1
2n+2
(x) set is dual Ramsey.
Proof. Let A be a
1
2n+2
(x) set and a
1
2n+2
(x) expression describing A,
i.e.,
y (y A [y]).
By Fact 2.1 we nd a M

-condition ', X` L[T

x
2n+1
] such that
either ', X` (pc(X
G
)) or ', X` (pc(X
G
)),
where X
G
is the name for a dual Mathias generic partition.
Without loss of generality, we assume the former. By our assumption, we
actually have a generic partition Z over L[T
x
2n+1
] with ', X` G
Z
, where
G
Z
is the lter associated to Z, i.e.,
't, Y ` G
Z
Z (t, Y )

.
This means that Z (X)

. Now by 2.1 (homogeneity of M

) again, every
element Z

of (Z)

is also M

-generic over L[T

x
2n+1
]. Since G
Z
G
Z
, we
still have ', X` G
Z
. Consequently, we have L[T
x
2n+1
][Z

] [= [pc(Z

)].
But was absolute for models containing T
x
2n+1
by Lemma 6.5, hence we
have V [= [pc(Z

)].
Summing up, we have found a partition Z such that pc(Z

) : Z

_ Z
A. This is exactly what we had to show by Observation 2.5.
Note that this type of argument probably will not work if you replace
dual Ramsey by completely dual Ramsey. What you would have to do
is to relativize the argument to arbitrary partitions W V. But at least
this does not work in the classical case: Brendle has shown in [Br00-1] that
in any model containing one Mathias real over a ground model N, there is
an Ellentuck neighbourhood that doesnt contain any Mathias reals over N.
Another useful comment about Theorem 6.6 is that if you look at the
case n = 0 you get a third proof of the consistency of Every
1
2
set is dual
Ramsey:
Corollary 6.7. Suppose that for each real x []

there is a dual Mathias

generic partition over L[x]. Then every
1
2
set is dual Ramsey.
Proof. Immediate from Theorem 6.6, keeping in mind that T
1
L by
[KeMo78, 9C], as mentioned in Subsection 2.7.
140 L. HALBEISEN AND B. L

OWE
This particularly generic version of proving the consistency of properties
of
1
2
sets should be compared to analogous results for Random forcing,
Cohen forcing and Hechler forcing.
17
We now move on to use Lemma 6.4 and Theorem 6.6 to get that Deter-
minacy implies the dual Ramsey property:
Corollary 6.8. Assume Det(
1
2n+2
). Then every
1
2n+2
set is dual Ram-
sey.
Proof. By Lemma 6.4, we get that for any reals x, y []

, the set of reals

in L[T
x
2n+1
, y] is countable.
By the same argument that is used to show that x(
L[x]
1
<
V
1
) im-
plies that
V
1
is strongly inaccessible in every L[x], we get that {(()

)
L[T
x
2n+1
, y] is countable for arbitrary choices of x and y []

.
Thus there are dual Mathias generic partitions over each L[T
x
2n+1
, y], in
particular over each L[T
x
2n+1
], and we can use Theorem 6.6 to prove the
claim.
7. Appendix: Game-lters have the segment-colouring-property.
Let F ()

be an ultralter. Associated with F we dene the game (

F
as
follows. This type of game, which is the Choquet-game with respect to the
dual Ellentuck topology (cf. [Ke95, 8.C]), was rst suggested by Kastanas
in [Kas83].
I 't
0
, Y
0
` 't
1
, Y
1
` 't
2
, Y
2
`
. . .
II 'X
0
` 'X
1
` 'X
2
`
All the moves X
n
of player II must be elements of the ultralter F and all
the moves 't
n
, Y
n
` of player I plays must be such that Y
n
F and (t

n
, Y
n
)

is a dual Ellentuck neighbourhood. Further, the nth move X

n
of player II is
such that X
n
(t

n
, Y
n
)

and then player I plays t

n+1
such that t

n
t

n+1
_
X
n
and [t
n+1
[ = [t
n
[ +1 = [t
0
[ +n+1. Player I wins if and only if the unique
Y with t
n
Y (for all n) is not in F.
An ultralter F is a game lter if and only if player I has no winning
strategy in the game (
F
.
18
In the following we outline the proof that game lters are also scp-lters.
The crucial point will be to show the Preliminary Lemma 7.1, which is in
17
Cf. [BrLo99]. Note that in most cases the existence of generics doesnt give more
than regularity at the
1
2
level, and something more than mere existence is needed for
the
1
2
level.
18
For the existence of game lters see [Halb98-1], where one can nd also some results
concerning dual Mathias forcing restricted to such lters.
APPROACHING THE DUAL RAMSEY PROPERTY 141
fact Carlsons Lemma (cf. [CaSi84, Lemma 2.4]) restricted to game lters.
But rst we have to give some notations.
Let s, t (N) be such that s _ t, [s[ = n and [t[ = m. For k with
k mn let
(t)
k
s
:= u (N) : dom(u) = dom(t) s u _ t [u[ = [s[ +k.
For s t _ X, let
(t, X)
k
s
:= u (N) : t u

_ X [u[ = [s[ +k,

and let (X)
k
s
:= (s, X)
(n+k)
.
We have chosen this notation following [CaSi84, Denition 2.1], where
one can consider s as an alphabet of cardinality n.
For the remainder of this section, let F be an arbitrary but xed game
lter.
Preliminary Lemma 7.1. Let s X F and : (X)
0
s
l, then there
exists a Y (s, X)

F such that (Y )
0
s
is constant.
Following the ideas of the proof of Theorem 6.3 of [CaSi84], the proof of
the Preliminary Lemma will be given in a sequence of lemmas. We start by
stating the well-known HalesJewett Theorem in our notation.
Hales-Jewett Theorem 7.2. Let s (N). For all d , there is an
h such that for any t (N) with s t and [t[ = [s[ + h, and for any
colouring : (t)
0
s
d, there is a u (t)
1
s
such that (u)
0
s
is monochromatic.
The number h in the HalesJewett Theorem depends only on the number
d and the size of [s[. Let HJ(d, [s[) denote the smallest number h which
veries the HalesJewett Theorem.
Let s, t (N) and X F be such that s t _ X. A set K (N) is called
dense in (t, X)

, if for all Y (t, X)

F, there is a u with t u

_ Y
which belongs to K. A set D ()
k
s
is called k-dense in (t, X)
k
s
, if for
all Y (t, X)

F, we have (t, Y )
k
s
D = .
Lemma 7.3. Let s t X F and assume that D ()
0
s
is 0-dense in
(X)
0
s
. Further assume that K = u : t u (u)
0
s
D = is dense in
some (t, Z)

, where Z (t, X)

F. Then there is an s

_ Z with t s
such that for all v

_ Z with s v we have (v)

0
s
D = .
Proof. We shall dene a strategy for player I in the game (
F
, such that
player I can follow this strategy just in the case when Lemma 7.3 fails. This
means that for every s

_ Z with t s there is a v

_ Z with s v such
that (v)
0
s
D = .
Let t

0
_ Z be such that t t
0
, [t[ = [t
0
[ and (t
0
)
0
s
D = . Further put
Y
0
= t
0
Z and player I plays 't
0
, Y
0
`. Assume 't
m
, Y
m
` is the mth move
of player I and player II replies with 'X
m
`. If the lemma fails with s = t

m
,
142 L. HALBEISEN AND B. L

OWE
player I can play 't
m+1
, Y
m+1
`, according to the rules of the game, such that
(t
m+1
)
0
s
D = .
Since F is a game lter, the strategy of player I is not a winning strategy
and the unique Y (Z)

such that t
m
Y (for all m ) belongs to F.
Take an arbitrary u with t u

_ Y . For such a u we nd a t
n
Y such
that u _ t
n
and dom(u) = dom(t
n
). By the strategy of player I we have
(t
n
)
0
s
D = and therefore (u)
0
s
D = . But this is a contradiction to the
assumption that K is dense in (t, Z)

. Hence, player I cannot follow this

strategy, which completes the proof.
Lemma 7.4. Suppose s t X F, D is 0-dense in (X)
0
s
and K = u :
t u (u)
0
s
D = is dense in (t, Z)

, where Z (t, X)

F. Then
there is a

t

_ Z with t

t and [

t[ = [t[ + 1 such that (

t)
0
s
D.
Proof. Let s be as in the Lemma 7.3, and let d := [( s)
0
s
[. By the Hales
Jewett Theorem, let h := HJ(d, [s[). Pick v (N) such that s v

_ X and
[v[ = [ s[ +h. Let s
i
: s
i
( s)
0
s
i d be an enumeration of the elements
of ( s)
0
s
. We colour (v)
0
s
by stipulating (u) = i if and only if s
i
uu D.
By the choice of n, there are

t (v)
1
s
such that (

t)
0
s
is monochromatic, and
therefore, (

t)
0
s
D. Thus, we have found a

t with t

t and [

t[ = [t[ + 1
such that (

t)
0
s
D.
Lemma 7.5. Suppose s X F and D is 0-dense in (X)
0
s
. Then there
are t (X)
1
s
and Y (t, X)

F such that u : t u (u)

0
s
D is
1-dense in (t, Y )
1
s
.
Proof. In a similar way as above we can dene a strategy for player I in the
game (
F
, such that player I can follow this strategy only if Lemma 7.4 fails.
But if Lemma 7.4 is wrong, this would yield because F is a game lter a
contradiction (cf. [CaSi84, Lemma 6.5]).
Notice that in Lemma 7.5 we did not require that the r (N) for which
we have r

t (X)
1
s
belongs to D. This we do in
Lemma 7.6. Suppose s X F and D is 0-dense in (X)
0
s
. Then there
are r

t (X)
1
s
and Y (t, X)

F such that u : t u (u)

0
s
D is
1-dense in (t, Y )
1
s
and r D.
Proof. Let Y
0
(t
0
, X)

F be as in the conclusion of Lemma 7.5. Thus

D
0
:= u : t
0
u (u)
0
s
D is 1-dense in (t
0
, Y
0
)
1
s
. Using Lemma 7.5,
player I can play 't
m
, Y
m
` at the mth move such that D
m
:= u : t
m

u (u)
0
s
D is (m+ 1)-dense in (t
m
, Y
m
)
(m+1)
s
.
Because player I has no winning strategy, the unique Y (s, X)

such
that t
m
Y (for all m) belongs to F, and because D is 0-dense in (X)
0
s
,
there is an r (Y )
0
s
which belongs to D. Let

t (t
m
)
1
s
be such that r

t.
Since (

t)
0
s
(t
m
)
0
s
and because D
m
is (m + 1)-dense in (t
m
, Y
m
)
(m+1)
s
we
APPROACHING THE DUAL RAMSEY PROPERTY 143
get u :

t u (u)
0
s
D is 1-dense in (

t, Y
m
)
1
s
. Hence, we have found an
r

such that u : r

u (u)
0
s
D is 1-dense in (r

, Y )
1
s
and r D.
Now we can go back to the
Proof of the Preliminary Lemma. Let s X F. We have to show that
for any colouring : (X)
0
s
l, there is a Y (s, X)

F such that (Y )
0
s
is monochromatic.
It is easy to see that at least one of the colours is 0-dense in (X)
0
s
, say
j and let D := t (X)
0
s
: (t) = j. Now we can prove the Prelimi-
nary Lemma in almost the same way as Lemma 7.6, the only dierence is
that player I uses now Lemma 7.6 to construct the mth move, instead of
Lemma 7.5.
Finally we get the main result of this section.
Proposition 7.7. Each game lter is also an scp-lter.
Proof. We have to show that for any colouring : (s, )
(|s|+k)
r, where
r and k are positive natural numbers and s (N), there is an X F such
that s X and (s, X)
(|s|+k)
is monochromatic.
Following the proof of [Halb, Theorem] and using the Preliminary
Lemma, it is not hard to dene a strategy for player I in such a way that
if player I follows this strategy, then for the resulting partition X which
must belong to F, since F is a game lter we get s X and (s, X)
(|s|+k)
is monochromatic.
We have seen that every game lter has the segment-colouring property.
It seems that the reverse implication is unlikely, since a strategy for player I
cannot be encoded by a real number, which makes it hard (if not impossible)
to prove that CH implies the existence of game lters. But on the other hand
we know that scp-lters always exist if we assume CH.
References
[AdMo68] J.W. Addison and Y.N. Moschovakis, Some consequences of the axiom of
denable determinateness, Proceedings of the National Academy of Sciences
U.S.A., 59 (1968), 708-712, MR 36 #4979, Zbl 186.25302.
[BaJu95] T. Bartoszy nski and H. Judah, Set Theory: On the Structure of the Real
Line, A. K. Peters, Wellesley, 1995, MR 96k:03002, Zbl 834.04001.
[BeKe84] H.S. Becker and A.S. Kechris, Sets of ordinals constructible from trees
and the third Victoria Delno Problem, in Axiomatic Set Theory
(J.E. Baumgartner, D.A. Martin, S. Shelah, Eds.), Contemporary Math-
ematics, 31, American Mathematical Society, (1984), 13-29, MR 86a:03051,
Zbl 629.03027.
144 L. HALBEISEN AND B. L

OWE
[Br00-1] J. Brendle, How small can the set of generics be?, in Logic Colloquium
98, Proceedings of the 1998 Association for Symbolic Logic European
Summer Meeting, Prague, Czech Republic, 1998 (S.R. Buss, P. Hajek,
P. Pudlak, Eds.), Lecture Notes in Logic, 13, Springer, (2000), 92-109,
MR 2001a:03099.
[Br00-2] , Martins axiom and the dual distributivity number, Mathematical
Logic Quarterly, 46(2) (2000), 241-248, MR 2001c:03084.
[BrLo99] J. Brendle and B. Lowe, Solovay-type characterizations of forcing alge-
bras, Journal of Symbolic Logic, 64 (1999), 1307-1323, MR 1,779764,
Zbl 945.03071.
[CaSi84] T.J. Carlson and S.G. Simpson, A dual form of Ramseys Theorem, Ad-
vances in Mathematics, 53 (1984), 265-290, MR 85h:04002, Zbl 564.05005.
[El74] E. Ellentuck, A new proof that analytic sets are Ramsey, Journal of Symbolic
Logic, 39 (1974), 163-165, MR 50 #1887, Zbl 292.02054.
[Halb98-1] L. Halbeisen, Symmetries between two Ramsey poperties, Archive for Math-
ematical Logic, 37 (1998), 241-260, MR 99i:03055, Zbl 937.03057.
[Halb98-2] , On shattering, splitting and reaping partitions, Mathematical Logic
Quarterly, 44 (1998), 123-134, MR 98k:03102, Zbl 896.03036.
[Halb] , A Ramsey type theorem and its corollaries, preprint.
[HalbJu96] L. Halbeisen and H. Judah, Mathias absoluteness and the Ramsey prop-
erty, Journal of Symbolic Logic, 61 (1996), 177-193, MR 97c:03119,
Zbl 851.03013.
[HalbLo] L. Halbeisen and B. Lowe, Ultralter spaces on the semilattice of partitions,
Topology and its Applications, to appear.
[HalJe63] A.W. Hales and R.I. Jewett, Regularity and positional games, Transactions
of the American Mathematical Society, 106 (1963), 222-229, MR 26 #1265,
Zbl 113.14802.
[HarKe81] L.A. Harrington and A.S. Kechris, On the determinacy of games on ordi-
nals, Annals of Mathematical Logic, 20 (1981), 109-154, MR 83c:03044,
Zbl 489.03018.
[Je78] T. Jech, Set Theory, Academic Press, San Diego, 1978, MR 80a:03062,
Zbl 419.03028.
[JuSh89] H. Judah and S. Shelah,
1
2
-sets of reals, Annals of Pure and Applied Logic,
42 (1989), 207-223.
[JuSh93] ,
1
3
-sets of reals, Journal of Symbolic Logic, 58 (1993), 72-80,
MR 94c:03067, Zbl 786.03036.
[Kan94] A. Kanamori, The Higher Innite, Perspectives in Mathematical Logic,
Springer-Verlag, Berlin Heidelberg, 1994, MR 96k:03125, Zbl 813.03034.
[Kas83] I.G. Kastanas, On the Ramsey property for sets of reals, Journal of Symbolic
Logic, 48 (1983), 1035-1045, MR 85j:03080, Zbl 527.90101.
[Ke95] A.S. Kechris, Classical Descriptive Set Theory, Graduate Texts in
Mathematics, 156, Springer-Verlag, New York, 1995, MR 96e:03057,
Zbl 819.04002.
[KeMo72] A.S. Kechris and Y.N. Moschovakis, Two Theorems about Projective
Sets, Israel Journal of Mathematics, 12 (1972), 391-399, MR 48 #1900,
Zbl 257.02034.
APPROACHING THE DUAL RAMSEY PROPERTY 145
[KeMo78] , Notes on the Theory of Scales, in Cabal Seminar 76-77,
Proceedings, Caltech-UCLA Logic Seminar, 1976-77 (A.S. Kechris and
Y.N. Moschovakis, Eds.), Lecture Notes in Mathematics, 689, Springer-
Verlag, Berlin, (1978), 1-53, MR 83b:03059, Zbl 397.032.
[Ku83] K. Kunen, Set Theory, an Introduction to Independence Proofs, Studies in
Logic and the Foundations of Mathematics, 102, North Holland, Amst-
erdam, 1983, Zbl 534.03026.
[Mar68] D.A. Martin, The axiom of determinateness and reduction principles in
the analytic hierarchy, Bulletin of the American Mathematical Society, 74
(1968), 687-689, MR 37 #2607, Zbl 165.31605.
[Mat77] A.R.D. Mathias, Happy families Annals of Mathematical Logic, 12 (1977)
59-111, MR 58 #10462, Zbl 369.02041.
[Mo71] Y.N. Moschovakis, Uniformization in a playful universe, Bulletin of the
American Mathematical Society, 77 (1971), 731-736, MR 44 #2609,
Zbl 232.04002.
[Mo80] , Descriptive Set Theory Studies in Logic and the Foundations
of Mathematics, 100, North-Holland, Amsterdam, 1980, MR 82e:03002,
Zbl 433.03025.
[Pl86] S. Plewik, On completely Ramsey sets, Fundamenta Mathematicae, 127
(1986), 127-132, MR 88d:04005.
[Rai84] J. Raisonnier, A mathematical proof of S. Shelahs Theorem on the Measure
Problem and related results, Israel Journal of Mathematics, 48 (1984), 48-56,
MR 86g:03082b, Zbl 596.03056.
[Ram29] F.P. Ramsey, On a problem of formal logic, Proceedings of the London
Mathematical Society, Ser. II, 30 (1929), 264-286.
[St95] J.R. Steel, Projectively well-ordered inner models, Annals of Pure and Ap-
plied Logic, 74 (1995), 77-104, MR 96h:03089, Zbl 821.03023.
Received September 14, 1999. The rst author wishes to thank the Swiss National Science
Foundation for its support during the time at U.C. Berkeley in which the research for
this paper has been done. The second author wishes to thank the Studienstiftung des
deutschen Volkes for the funding of his travels to Boise ID and Las Vegas NV, and in
particular Gunter Fuchs (Berlin).
Queens University Belfast
Belfast BT7 1NN
Northern Ireland
E-mail address: halbeis@qub.ac.uk
Mathematisches Institut
Rheinische Friedrich-Wilhelms-Universit

at Bonn
Beringstrae 6
53115 Bonn
Germany
E-mail address: loewe@math.uni-bonn.de